What Happens When an AI Trains on Its Own Outputs For 12 Years With No Human Input?
中文摘要
本文探讨了“模型崩溃”现象，即AI若在缺乏人类输入的情况下，仅利用自身生成的合成数据进行长期训练，会导致模型性能退化和信息丢失。
English Summary
This article explores "model collapse," where AI models trained solely on their own synthetic outputs without human input suffer from progressive information degradation and loss of accuracy.
原文节选
Model Collapse is a real and documented AI phenomenon. A thought experiment asks what lives at the mathematical foundation of all human… Continue reading on Medium »